AI Act · Deepfakes · Art. 50
Deepfakes and transparency
A convincing fake video is cheap to make and travels fast. The AI Act answers with a rule that is simple to state and demanding to apply: AI-generated content must be recognisable as such. This page explains who must do what — and why the real stake is trust, not just individual videos.
01In brief
The AI Act does not ban deepfakes: it requires transparency. Systems that generate synthetic content must make it machine-detectable as artificial, and whoever uses a deepfake must disclose that the content was artificially generated or manipulated (Art. 50). The problem the law targets is not the technology — it is deception.
02What Article 50 requires
- Providers: systems generating synthetic audio, images, video or text must mark outputs in a machine-readable format so they are detectable as artificially generated.
- Deployers: whoever uses a system to create or manipulate a deepfake must disclose that the content is artificial. For clearly artistic, satirical or creative content the rule allows lighter disclosure — an area of interpretation, not a free pass.
- Chatbots: people must know when they are talking to a machine — same logic, different channel (transparency obligations).
03Why the label matters
An unlabelled deepfake does not harm only the person imitated: it erodes trust in everything we see. If any video might be fake, authentic videos become deniable too — the so-called "liar's dividend". Mandatory transparency protects the information ecosystem before it protects individual victims, which is why violating it in the game damages the whole city's trust indicators.
04In the game: "The synthetic city"
In the case The synthetic city you inspect a platform generating content with no marking at all: you must decide whether the problem is the generating model, the way it is used, or both — and who is responsible. It is the fastest way to grasp the provider/deployer distinction under Article 50. Play the case, then compare your decision with the real rule.
05What this page does NOT cover
Defamation, copyright, image rights and offences under national law sit outside the AI Act and outside this page: a labelled deepfake can still be unlawful for other reasons. For a concrete case you need a professional, not an educational page.
06Official sources and transparency
- Content responsibility: the NO AI ACT project (Matteo Angeloni); internal editorial review, no third-party legal review.
- First published: July 2026 · Last reviewed: July 2026 (site v2.3.0).
- Official sources: Regulation (EU) 2024/1689 on EUR-Lex (Art. 50); EU regulatory framework for AI (European Commission); European AI Office.
- Related game cases: "The synthetic city" (generated content), "The desk that always answers" (chatbot).
- Related teacher resources: classroom activities (transparency hunt), classroom lab guide.
- Disclaimer: this page is an educational simplification, not legal advice.